11

For integers $n\geq 1$, $\operatorname{rad}(n)$ denotes the radical of an integer, see in Wikipedia this definition $$\operatorname{rad}(n)=\prod_{p\mid n}p,$$ if $n>1$ with factorization $n=\prod_{p\mid n}p^{e_p}$, and defining $\operatorname{rad}(1)=1$.

I'm inspired in a similar conjecture due to Amdeberhan, and Medina and Moll.

Question. I believe that the following conjecture holds. Can you prove or refute it?

There exists an integer $N$ such that $\forall n>N$ $$\prod_{k=1}^n(1+(\operatorname{rad}(k))^2)$$ is not a square.

Many thanks.

References:

Amdeberhan, and Medina and Moll, Arithmetical properties of a sequence arising from an arctangent sum, J. Number Theory (2007).

  • 1
    If some user is interested in the (cited) genuine conjecure, was solved by Cilleruelo, Squares in $(1^2+1)\cdots(n^2+1)$, Journal of Number Theory, Vol. 128, Issue 8, (2008), Pages 2488-2491. –  May 28 '17 at 15:52
  • I misread your statement. What is it supposed to be, where does it come from ? You say it is similar to something then why don't you explain it ? – reuns Jun 26 '17 at 22:19
  • Many thanks for your attention @user1952009 My question is similar than squares in $(1^2+1)\cdots(n^2+1)$, but my question has no more mathematical meaning than a curiosity. That is the original conjecture by Amdeberhan, and Medina and Moll has mathematical meaning from an original problem, but not mine, my question is only a simple curiosity insipired in it. –  Jun 26 '17 at 22:53
  • 1
    You might want to search for $a$ square-free such that $a^2+1$ is prime. In that case if $a^2+1$ divides $(a+b)^2+1=a^2+1+b^2+2ab$ then it divides $b$ or $2a+b$. So $a^2+1$ doesn't divide $rad(k)^2+1$ for $k<a$, nor for $rad(k)−a<a^2+1$ ie. nor for $k\le a^2+a$. Thus $\prod_{k=1}^{a+m} (rad(k)^2+1)$ is not a square for $0 \le m \le a^2+a$ – reuns Jun 26 '17 at 22:57
  • Many thanks @user1952009 as I've said to you a lot of times I like much your mathematics in this site, you are generous and your mathematics are very very good. I hope don't disturb to you with my commments. Good week. –  Jun 28 '17 at 16:02

1 Answers1

6

Important Note: The argument detailed below proves the desired theorem if the quantity $$C_{\text{rad}}=\frac{6}{\pi^2}\sum_{l=1}^\infty \frac{d_l}{l(l+1)}$$ satisfies $C_{\text{rad}}>\frac{1}{2}$, where $d_l$ is the relative density of squarefree integers $k$ in the interval $\frac{n}{l+1}<k\leq \frac{n}{l}$ with $\text{num_rad}_n(k)=\text{odd}$, where $\text{num_rad}_n$ is defined below. Currently I can prove that $$0.46684<C_{\text{rad}}<0.522,\ \ \ \ \ \ \ \ \ \ (1)$$ but I do not know how to evaluate this series exactly. If $C_{\text{rad}}<\frac{1}{2}$, then this proof technique fails to achieve the desired result.


For squarefree integers $k$, define the function $$\text{num_rad}_{n}(k)=\#\left\{ j\leq n:\ \text{rad}(j)=\text{rad}(k)\right\}.$$ Since squarefree integers have unique radicals, it follows that $$\prod_{k=1}^{n}\left(1+(\text{rad}(k))^{2}\right)=\prod_{\begin{array}{c} k\leq n\\ k\text{ squarefree} \end{array}}\left(1+(\text{rad}(k))^{2}\right)^{\text{num_rad}_{n}(k)}.$$ Split the squarefree integers below $n$ into two subsets, $S_{n}^{\text{odd}}$ and $S_{n}^{\text{even}},$ defined by $$S_{n}^{\text{odd}}=\left\{ k\leq n\text{ squarefree}:\ \text{num_rad}_{n}(k)\text{ is odd}\right\}$$ and $$S_{n}^{\text{even}}=\left\{ k\leq n:\ \text{num_rad}_{n}(k)\text{ is even}\right\}.$$ Then $$\prod_{k=1}^{n}\left(1+(\text{rad}(k))^{2}\right)=K^{2}\prod_{k\in S_{n}^{\text{odd}}}\left(1+(\text{rad}(k))^{2}\right)$$ for some integer $K$, where $\text{rad}(k)\neq\text{rad}(j)$ for any distinct $j,k\in S_{n}^{\text{odd}}$. Our goal is to lower bound the size of the squarefree part of the quantity

$$Q(n)=\prod_{k\in S_{n}^{\text{odd}}}\left(1+(\text{rad}(k))^{2}\right),$$

and we do so in the following way:

  • Step 1: Upper bound the size of the squared part using an argument similar to the answer here: On prime factors of $n^2+1$
  • Step 2: Lower bound the size of $Q(n)$.

Unfortunately the lower bound may not be large enough, and relies on the constant $C_{\text{rad}}$ being greater than $1/2$.


Bounding the Squareful Part

Proposition 1: Suppose that $Q(n)=M^2\cdot R$ where $R$ is squarefree. Then $$\log M\lesssim \frac{n\log n}{2}$$

Proof: We first establish that if $p|M^2$ then $p\leq 2n$. If $p$ divides only one term $(1+\text{rad}(k)^2)$ then we must have $p^2$ dividing that term, and so $p\leq 2\text{rad}(k)\leq 2n$. If $p$ divides two different terms, $(1+\text{rad}(k)^2)$, $(1+\text{rad}(j)^2)$ then $p$ divides $(\text{rad}(k)-\text{rad}(j))(\text{rad}(j)+\text{rad}(k))$ and so once again $p\leq 2n$. It is a theorem of Fermat that if $p$|(m^2+1) for an integer $m$, then $p\equiv 1\pmod{4}$. This congruence fact is critical, as it allows us to gain a factor of $2$ in the exponent over the trivial bound.

Thus, we may write $$M^2=\prod_{\begin{array}{c} p<2n\\ p\equiv1\ (4) \end{array}}p^{\alpha_{p}}$$ where each $\alpha_{p}$ is even. Now, among the integers $1,\dots,p^k$, there can be at most two for which $p^k|x^2+1$, and so for each $p\equiv 1\pmod{4}$ $$\alpha_{p}\leq\sum_{j\leq\log_{p}(n^{2}+1)}2\lceil\frac{n}{p^{j}}\rceil\leq2\log_{p}(n^{2}+1)+\frac{2n}{p-1}.$$ This yields the upper bound $$\log M^2\leq\sum_{\begin{array}{c} p<2n\\ p\equiv1\ (4) \end{array}}\left(\frac{2n\log p}{p-1}+2\log(n^{2}+1)\right),$$ $$\leq 2\pi(n,4,1)\log(n^2+1)+2n\sum_{\begin{array}{c} p<2n\\ p\equiv1\ (4) \end{array}}\frac{\log p}{p-1}.$$ Asymptotically, by the prime number theorem for arithmetic progressions, it follows that

$$\log M^2 \lesssim n\log n.$$


Lower Bounding $\log Q(n)$

Let $k$ be a squarefree number. Then we want to understand how many integers below $n$ have $k$ as their radical, that is the number of divisors $d|k$ such that $d\cdot k\leq n.$ Let $M_{n}^{l}=\left\{ k\text{ squarefree}:\ \frac{n}{l+1}\leq k\leq\frac{n}{l}\right\},$ where $l\geq1$ is an integer. We split into these ranges since $$d\cdot k=\begin{cases} \leq n & \text{if }d\leq l\\ >n & \text{if }d>l \end{cases},$$ which means that we are working a truncated divisor function. In what follows, we will attempt to calculate the density $d_{l}$ of squarefree integers $k$ in $M_{n}^{l}$ that also lie in $S_{n}^{\text{odd}}$, that is that have $\text{num\_rad}_{n}(k)$ odd. With this notation, it follows that $$\frac{|S_{n}^{\text{odd}}|}{n}\sim\frac{6}{\pi^{2}}\cdot\left(\sum_{l=1}^{\infty}\frac{d_{l}}{l(l+1)}\right)=C_{\text{rad}},$$ where the $\frac{6}{\pi^{2}}$ factor appears since it is the density of the squarefree integers.

Taking the product over $l$ of the smallest element in $M_l$ to the power of the number of elements in this interval, interval, we find that $$\log(Q(l))\geq \frac{6}{\pi^2} \sum_{l=1}^\infty \frac{d_l}{l(l+1)}\log\frac{n}{l+1}=C_{\text{rad}} n\log n+O(n),$$ and so we have proven the following proposition:

Proposition 2: We have that $$\log(Q(n))\geq C_{\text{rad}} n\log n+O(n).$$

Combining proposition $1$ and $2$ we immediately deduce:

Theorem 1: If $C_{\text{rad}}>\frac{1}{2}$, then there exists a large constant $N_0$ such that for all $n>N_0$ $$\prod_{k=1}^n \left(1+\left(\text{rad}(k)\right)^2\right)$$ is never a square.


Calculations: Calculating $d_l$ for $l=1,2,3,4$

When $l=1,$ all the integers in $M_{n}^{l}$ lie in $S_{n}^{\text{odd}}$ since $d=1$ is the only allowable divisor, and hence $$d_{1}=1.$$ When $l=2,$ then those squarefree integers divisible by $2$ must be excluded. This amounts to multiplying by $\left(1-\frac{1}{2}\right),$ but since the squarefree integers already excluded multiples of $4,$ we must divide by $\left(1-\frac{1}{2^{2}}\right),$ and so $$d_{2}=\frac{\left(1-\frac{1}{2}\right)}{\left(1-\frac{1}{2^{2}}\right)}=\frac{2}{3}.$$ When $l=3,$ we must avoid factors of $2,$ factors of $3,$ but not factors of $2$ and $3,$ since in that case we have an odd number of divisors. Thus $$d_{3}=\frac{\left(\left(1-\frac{1}{2}\right)\left(1-\frac{1}{3}\right)+\frac{1}{2\cdot3}\left(1-\frac{1}{2}\right)\left(1-\frac{1}{3}\right)\right)}{\left(1-\frac{1}{2^{2}}\right)\left(1-\frac{1}{3^{2}}\right)}=\left(\frac{7}{12}\right).$$ When $l=4,$ we must only avoid factors of $3,$ and so $$d_{4}=\frac{\left(1-\frac{1}{3}\right)}{\left(1-\frac{1}{3^{2}}\right)}=\frac{3}{4}.$$ In general, we will have that $$d_l=\prod_{p\leq l}\left(1-\frac{1}{p}\right)\cdot \prod_{p\leq l}\left(1-\frac{1}{p^2}\right)^{-2}\cdot \left(\sum_{n}\frac{f_l(n)}{n}\right)$$ for a very specific function $f_l(n)$ taking values $0$ or $1$, and which is always $0$ if $p>l$ divides $n$ or if $n$ is not squarefree, and is $1$ if $n=1$. Cleaning this up a little, $$d_l=\prod_{p\leq l}\left(1+\frac{1}{p}\right)^{-1}\left(\sum_{n=1}^\infty\frac{f_l(n)}{n}\right).$$ In particular, for primes $q$, we have that $$f_l(q)=\sum_{q^k\leq l}(-1)^{k-1},$$ and in general it seems fairly messy.

Explicit Calculations:

Using these numbers, we obtain the bounds stated in equation $(1)$.

Eric Naslund
  • 72,099
  • Many thanks I am going to read your great answer. And also thanks for sharing on this site these calculations of your result. –  Jun 28 '17 at 09:17
  • 1
    @user243301: I have made a major update to the answer. I am actually not entirely sure this proves the theorem, since I am not sure if the quantity I mentioned is indeed larger than $1/2$. It can be written as an infinite series, and at the moment I do not know how to evaluate that infinite series. One possibility is to use a computer to evaluate the first $1000$ or so terms, and check whether it's greater than $1/2$, but calculating the individual terms is quite messy as you can see form the calculations section, and a bit more thought needs to be put into this. – Eric Naslund Jun 28 '17 at 19:54
  • Your calculations exceed my level of understanding and knowledge. I am going to try to understand the more possible. We can wait constructive feedback from other mathematicians. Any case you are very generous with these calculations. –  Jun 28 '17 at 21:22
  • 1
    @user243301: In that case, I recommend trying to understand the proof given in this answer first: https://math.stackexchange.com/questions/589344/on-prime-factors-of-n21/591069#591069 (In fact, you can actually improve it to $2n\log n$ by noting that primes $p>2n$ can only divide the product once.) That method is known as the "Chebyshev Method". In the event that $C_{\text{rad}}<1/2$, the theorem is likely still true, but would unfortunately require stronger tools, like those used by Iwaniec and Deshoullier 1982. – Eric Naslund Jun 28 '17 at 21:36
  • Many thanks one more time. And of course, your work is excellent. –  Jun 28 '17 at 21:43